Donghao Yang

Hi, I am Donghao Yang (杨东浩), a Ph.D. student at the Software Engineering Institute, Beihang University.

My research interests lie in AI for Software Engineering (AI4SE), especially large language models and agentic systems for real-world software engineering tasks. I am interested in how AI can help developers understand, maintain, generate, and repair software, and how we can evaluate such systems in a rigorous and trustworthy way.

Currently, my work focuses on the following directions:

• LLM-based software maintenance and program repair, including how design rationales, issue discussions, and human-written solutions can guide automated repair.
• Knowledge-augmented code generation, especially for domain-specific and industrial software such as PLC Structured Text and Siemens SCL.
• Requirements engineering with LLMs, including semantic-role-based requirement completion and retrieval-augmented requirement understanding.
• Agent tooling and MCP ecosystems, with an emphasis on how tool design, feedback, schema quality, and error handling affect the performance of LLM agents.
• Evaluation methodology for AI4SE, including LLM-as-Judge calibration, rubric design, human agreement, and empirical validity.

A central theme of my research is to go beyond asking whether an AI system works. I care about why it works, when it fails, what evidence supports the result, and how its behavior changes in realistic software engineering workflows.

My recent projects include DRMiner / DRCodePilot for design-rationale-enhanced program repair, AutoPLC for knowledge-augmented industrial code generation, and empirical studies on LLM-based evaluation and agent-oriented tool usability.

I am always open to discussions on AI4SE, empirical software engineering, LLM agents, and research methodology.